Dierential Forms

MA 305
Kurt Bryan
Aesthetic pleasure needs no justication, because a life without such pleasure is one not
worth living. Dana Gioia, Can Poetry Matter?
Introduction
Weve hit the big three theorems of vector calculus: Greens, Stokes, and the Divergence
Theorem (although Greens theorem is really a pretty obvious special case of Stokes theorem). But its an amazing fact that all of these theorems are really special cases of an even
larger theorem that unies all that weve studied so far. This new theory also extends what
weve done to spaces of arbitrary (nite) dimension. It rests on the idea of dierential forms.
Dierential forms are hard to motivate right o the bat. It wont be immediately clear
how dierential forms are related to what weve done, but be patient. For now treat dierential forms as a new kind of mathematical object; were going to learn how to manipulate
themadd, multiply, dierentiate, and integrate. It will be quite abstract and formal at
rst, but youll soon see how they connect with what weve already done.
Manifolds
Youve already seen many examples of one and two dimensional manifolds. A one dimensional manifold C is just a curve. Such a curve can exist in any dimension and can
be parameterized as X(t) (Im using this instead of R), where t ranges over some interval
D = (a, b) in lR1 and
X(t) = (x1 (t), x2 (t), . . . , xn (t))
if the curve exists in n dimensions. You may recall that we required that our curves be
smooth, which meant that the curve had to have a proper parameterizationX(t) must be
dierentiable, injective, and X (t) = 0 for any t in D. Smooth one dimensional manifolds
are just smooth curves. Recall that the vector X (t) is tangent to the manifold C.
A smooth two dimensional manifold S is just a surface, of which weve seen many examples (in three dimensions.) Such a surface can be specied as X(u, v), where (u, v) ranges
over some appropriate subset D of lR2 . The function X(u, v) is of the form
X(u, v) = (x1 (u, v), x2 (u, v), . . . , xn (u, v)),
although weve mostly considered the case n = 3. There were some technicalities in dealing
with surfaces and integrating over themthe surface had to be orientable, and we did require
and X
be independent, i.e., neither a multiple of the other.
that the tangent vectors X
u
v
If this isnt true then our surface S might end up being only one-dimensional. Also, the
surface was not allowed to self-intersect.
Based on the above, its not much of a stretch to consider a k-dimensional manifold. Let
D be some bounded open region in lRk . Well use coordinates u = (u1 , u2 , . . . , uk ) on D.
Consider a function
X(u) = (X1 (u), . . . , Xn (u))
1

dened on D, which maps each point in D into a point in n dimensional space. Take
M = X(D), the image of D in lRn . We will say that M is a smooth k-manifold if
The function X is C 1 .
The function X is injective (distinct points in D go to distinct points in lRn .
The tangent vectors
X
=
uk

X1
Xn
,...,
uk
uk

are all linearly independent at all points in D. This means that the only solution to
c1

X1
Xn
+ + cn
=0
uk
uk

is c1 = c2 = = cn = 0. Since this is a semi-intuitive account of dierential forms

and manifolds, we wont harp too much on this last requirement.
Example 1: Let D be the open cube 0 < u1 , u2 , u3 < 1 in lR3 . Take the mapping
X from lR3 to lR5 to be dened by
X1 (u1 , u2 , u3 )
X2 (u1 , u2 , u3 )
X3 (u1 , u2 , u3 )
X4 (u1 , u2 , u3 )
X5 (u1 , u2 , u3 )

=
=
=
=
=

u1 u2
3u3
u21
u3 2u2
3.

The function X is obviously C 1 . Its a bit tedious (Maple makes it easier) but
you can check that no two points in D go to the same image point in lR5 , and the
tangent vectors are in fact all independent. The image M = X(D) is a smooth
3-manifold in lR5 .

Dierential Forms
Our discussions will take place in lRn . Well use (x1 , . . . , xn ) as coordinates, and write x
for the point (x1 , . . . , xn ). We will dene a 0-form to be simply a function on lRn , so
= f (x1 , . . . , xn ).
Thats all there is to say about 0-forms.
A basic or elementary 1-form in lRn is an expression like dxi , where 1 i n. More
generally, a 1-form is an expression like
= F1 (x) dx1 + F2 (x) dx2 + Fn (x) dxn

where the Fi are functions of x. Youve actually encountered 1-forms before, when we did
Greens theorem. Recall that Greens Theorem said
(
D

F2 F1

x
y

dx dy =
D

F1 (x, y) dx + F2 (x, y) dy.

The integrand on the right is an example of a 1-form.

A dierential 1-form is not a passive object, but in fact can be thought of as a kind of
function. The basic 1-form dxi accepts as input a single vector v and outputs vi , the ith
component of v, so
dxi (v) = vi .
A general 1-form = F1 (x) dx1 + Fn (x) dxn acts on a single input vector v as
(v) = F1 (x)v1 + Fn (x)vn .
You can add two 0-forms in the obvious way (as functions). You can similarly add two
1-forms, e.g.,
(x22 dx1 + ex1 dx2 ) + (2 dx1 x1 dx2 ) = (x22 + 2) dx1 + (ex1 x1 ) dx2 .
A basic dierential 2-form is an expression like
= dxi dxj
where 1 i, j n. The symbol denotes what is called the wedge or exterior product.
Dont worry too much about what it means (yet); in the end it will mean essentially just
dxi dxj , an object suitable to stick under a double integral.
Like 1-forms, 2-forms also act on vectors. A basic two form = dxi dxj accepts as
input TWO vectors v1 and v2 . The output is the determinant
[

(v1 , v2 ) = dxi dxj (v1 , v2 ) = det

dxi (v1 ) dxi (v2 )

dxj (v1 ) dxj (v2 )

Based on the properties for the determinant that weve seen, you can immediately conclude that
dxi dxj = dxj dxi
dxi dxi = 0
This raises an interesting question: How many dierent 2-forms are there in n dimensions?
If you count all possible combinations for 1 i, j n for dxi dxj you get n2 , but in fact n
of these are really zero. That leaves n2 n possible 2-forms, although in some sense there is
only half that many, for the second property above shows that, e.g., dx1 dx2 = dx2 dx1 .
There are thus only (n2 n)/2 half this many independent 2-forms.
A (general) dierential 2-form is an expression of the form
= F12 (x)dx1 dx2 + F13 (x)dx1 dx3 + Fn1,n (x)dxn1 dxn =

1i<jn

Fij (x)dxi dxj .

Notice that we can always assume that i < j in the sum above, for if we had j < i we could
simply swap the order of dxj dxi and replace it with dxi dxj .
You can probably guess how such a 2-form acts on two input vectors:
(v1 , v2 ) =

Fij (x)dxi dxj (v1 , v2 ) =

Fij (x)det

i,j>i

1i<jn

dxi (v1 ) dxi (v2 )

dxj (v1 ) dxj (v2 )

Example 2: Let denote the 2-form in three dimensions

= (x1 + 2x3 )dx1 dx2 + x2 dx1 dx3 .
Lets compute applied to the vectors v1 = (1, 3, 3), v2 = (1, 0, 7). You get
[

(v1 , v2 ) = (x1 + 2x3 )det

[

dx1 (v1 ) dx1 (v2 )

dx2 (v1 ) dx2 (v2 )
]

+ x2 det
[

1 1
1 1
= (x1 + 2x3 )det
+ x2 det
3 0
3 7
= 3(x1 + 2x3 ) + 10x2 = 3x1 + 10x2 + 6x3 .

dx1 (v1 ) dx1 (v2 )

dx3 (v1 ) dx3 (v2 )

Now were ready for the denition of a basic k-form: a basic k-form in n dimensions is
an expression of the form
= dxi1 dxi2 dxik
where 1 ij n for all j. Such a k-form accepts as input k vectors v1 , . . . , vk to give output

(v1 , . . . , vk ) = det

dxi1 (v1 ) dxi1 (v2 ) dxi1 (vk )

dxi2 (v1 ) dxi2 (v2 ) dxi2 (vk )
..
..
..
..
.
.
.
.
dxik (v1 ) dxik (v2 ) dxik (vk )

By using the properties of determinants that weve already deduced, its easy to see that
dxi1 dxij dxik dxim = dxi1 dxik dxij dxim (1)
dxi1 dxij dxij dxim = 0
(2)
Given the above two facts, its interesting to contemplate the question how many independent basic k-forms are there in n dimensions? Its not too hard to gure out, so Ill leave
it for you.
A (general) k-form is an expression of the form
=

Fi1 ,...,ik (x)dxi1 dxi2 dxik .

1i1 ,...,ik n

Such a k-form accepts k input vectors to produce

(v1 , . . . , vk ) =

Fi1 ,...,ik (x)dxi1 dxi2 dxik (v1 , . . . , vk ).

1i1 ,...,ik n

(3)

With this notation, its easy to see why a function is called a 0-formit doesnt accept any
input vectors.
Integrating Dierential Forms
Integrating dierential forms is easy. The general rule is that one integrates a k-form
over a k-manifold.
Lets rst look at how to integrate a 1-form over a 1-manifold. Accordingly, let M = X(t)
with a t b be a smooth 1-manifold in lRn and let be a 1-form dened in a neighborhood
of M . We use M to denote the integral of over M and dene this integral by

(X (t)) dt.

(4)

Example 3: Let M be the 1-manifold in lR3 dened by X(t) = (3t, t2 , 5 t) for

0 t 2. You can see that M is just a curve in three dimensions. Let be
the 1-form = 2x2 dx1 x1 x3 dx2 + dx3 . Using equation (4) and recalling how a
1-form acts on input vectors we obtain

=
M

((3, 2t, 1)) dt =

=
0

((2x2 )(3) x1 x3 (2t) 1) dt

(6t3 24t2 1) dt = 42.

Example 4: Suppose that F(x) = (F1 , F2 , . . . , Fn ) is a vector eld in lRn (so

each of the Fi are functions of x). Dene the dierential form
= F1 dx1 + F2 dx2 + + Fn dxn .
If we compute the integral of over some 1-manifold dened by M = R(t) =
(x1 (t), . . . , xn (t)) we get

=
M

a
b

=
a

(F1 dx1 + F2 dx2 + + Fn dxn )(R (t)) dt

(F1

xn
x1
+ + Fn
) dt =
F dR.
t
t
M

In other words, the integral of the 1-form over M is just the familiar line
integral of the vector eld F over the curve M .
One can integrate a 2-form over a 2-manifold M . Suppose that M is parameterized by
M = X(u), where u = (u1 , u2 ) ranges over D, a region in lR2 . The integral is dened by

=
M

(
D

X X
,
) du1 du2 .
u1 u2

Example 5: Let M be a 2-manifold in lR4 parameterized by

X(u) = (u1 , u1 u2 , 3 u1 + u1 u2 , 3u2 )
5

where u21 + u22 < 1. Take = x2 dx1 dx3 x4 dx3 dx4 . You can easily compute
that
X
= (1, 1, u2 1, 0)
u1
X
= (0, 1, u1 , 3)
u2
and that

X X
1
0
(
,
) = x2 det
x4 det
u2 1 u1
u1 u2
= x2 u1 + 3x4 (u2 1)
= u21 u1 u2 9u22 + 9u2 .
As a result

=
M

1u2

1
1

1u21

u2 1 u1
0
3

(u21 u1 u2 9u22 + 9u2 ) du2 du1 = 2.

Example 6: Consider a smooth 2-manifold M in lR3 parameterized as X(u) for

u in D lR2 . Let
= F1 dx2 dx3 + F2 dx1 dx3 + F3 dx1 dx2
be a 2-form dened in a neighborhood of M . Then

=
M

F1 (dx2 dx3 )(

F1 det

X2
u1
X3
u1

X X
X X
X X
,
) + F2 (dx1 dx3 )(
,
) + F3 (dx1 dx2 )(
,
)
u1 u2
u1 u2
u1 u2
X2
u2
X3
u2

+ F2 det

X1
u1
X3
u1

X1
u2
X3
u2

+ F3 det

X1
u1
X2
u1

X1
u2
X2
u2

])

du1 du2

X1 X3 X1 X3
X2 X3 X2 X3

) + F2 (

)
u1 u2
u2 u1
u1 u2
u2 u1
D
X1 X2 X1 X2
+F3 (

) du1 du2
u
u
u
u
1
2
2
1

F1 (

=
M

F dA.

In other words, the integral of over M is just the ux of the vector eld F =
(F1 , F2 , F3 ) over M . If the step between the last two equations looks mysterious,
all Ive done is use the fact that dA = (X/u1 ) (X/u2 ) and dotted this
with F; its the usual procedure for computing ux over a parameterized surface.
You can probably see how we should integrate a general k-form over a k-manifold. If the
manifold M is parameterized by X(u) for u D lRk , and if we have a general k-form
given by an equation like (3) then

=
M

1i1 ,...,ik n D

Fi1 ,...,ik (X(u))(dxi1 dxik )(

X
X
,...,
) du1 duk .
u1
uk

Example 7: Let M be a 3-manifold in lR4 parameterized as

X(u1 , u2 , u3 ) = (u2 u3 , u21 , 1 3u2 + u3 , u1 u2 )
for u21 + u22 + u23 < 1. Take
= x3 dx1 dx2 dx4 .
Its easy to compute that
X
= (0, 2u1 , 0, u2 )
u1
X
= (u3 , 0, 3, u1 )
u2
X
= (u2 , 0, 1, 0).
u3
Then

1
u3 u2
X X X

(
,
,
) = x3 det 2u1 0 0 = 2u21 u2 6u21 u22 + 2u21 u2 u3 .
u1 u2 u3
u2 u1 0
Finally, we nd that

=
M

(2u21 u2 6u21 u22 + 2u21 u2 u3 ) du1 du2 du3 =

8
.
35

The last triple integral is easier if you convert it to spherical.

Example 8: Let M be an open region in lRn and suppose that M is parameterized as X(u) where u D, where D is another region in lRn . In other words, M
is an n-manifold in n-dimensional space. Of course, we could very easily parameterize M by taking D = M and using X(u) = u, the identity map, but this isnt
necessary, and by not doing so youll see some important issues in integrating
dierential forms over manifolds.
Let = f (x) dx1 dx2 dxn (this is the most general n-form in lRn why?)
Then

X
X
=
f (X(u))(dx1 dxn )(
,...,
)
u1
un
M
D
X

n
1
X

u1
u1
.
..
..
=
f (X(u))det
.
.
..
du1 . . . dun
D
Xn
X1
un
un
=

f (x1 , . . . , xn ) dx1 . . . dxn

where the last equality follows from the change of variables formula in n dimensions (notice that the absolute value signs on the determinant are missing). The
integral of over M is PLUS OR MINUS the integral of f over M . It could be
either, depending on the parameterization.
7

The last example

raises an important issue that we wont entirely resolve. You might

hope the quantity M should depend only on M and , but this is true only up to a minus
sign. The problem is that any manifold which is orientable has two possible orientations, and
this is where the sign ambiguity arises. Youve already encountered this in line and surface
integrals:
if C is a curve in two or three dimensions and F is a vector eld, the quantity

C F dR is not well-dened. It depends on which direction you parameterize or traverse the

curve C; if you consider this integral physically, as work done, it clearly depends on whether
youre moving from point A to point B or the reverse. The sign ambiguity also arose in
doing ux integrals over a surface. Which direction you choose for the normal aects the
sign of the answer.

It is a general fact (that Ill let you check for yourself) that the quantity M does not
depend on how M is parameterized, except for the sign of the answer. Proving this really comes down to the change of variable formula for integral of n variables. The dierent
signs are a manifestation of how we chose to orient M (implicitly) when we parameterized it.
The Wedge Product
The wedge (or exterior product allows us to multiply dierential forms. Suppose that
f is a 0-form (a function) and is a k-form as given in equation (3). The wedge product of
f with is just

f =
f Fi1 ,...,ik (x)dxi1 dxi2 dxik .
i1 ,...,ik

Suppose that is an m-form,

=

Gj1 ,...,jm (x)dxj1 dxj2 dxjm .

j1 ,...,jm

The wedge product of with is the k + m form given by adjoining the two forms, as
=

Fi1 ,...,ik (x)Gj1 ,...,jm (x)dxi1 dxik dxj1 dxjm .

As complicated as this looks, its pretty easy in any specic example, and the rules (1) and
(2) make the result a lot smaller than you might think.
Example 9: Let = x2 dx1 dx3 + dx1 dx4 and = (x1 + 1)dx2 dx4 . Then
= x2 (x1 + 1) dx1 dx3 dx2 dx4 + (x1 + 1) dx1 dx4 dx2 dx4 .
However, by equation (2) the second term above (which contains two copies of
dx4 ) is zero. Also, by property (1) we can ip the dx2 and dx3 in the rst term,
which will introduce a minus sign, so
= x2 (x1 + 1) dx1 dx2 dx3 dx4 .
Its not hard to prove the following properties for the wedge product: If f is a function,
1 and 2 are k-forms, an m-form, and any form then
(1 + 2 ) = 1 + 2
8

(1 ) = 1 ( )
(f 1 ) = f (1 ) = 1 (f )
1 = (1)km 1

Only the last property is at all nonobvious, but a little experimentation will show you how
to prove it. In summary, this last property says that = UNLESS both forms are
of odd degree, in which case the sign ips.
The Exterior Derivative
Given that dierential forms can be integrated, it stands to reason that they can also be
dierentiated. The exterior derivative d is an operator that turns k-forms into k + 1-forms,
according to very simple rules:
The exterior derivative of a 0-form f (x) in lRn (recall, f is just a function) is
df =

f
f
dx1 + +
dxn .
x1
xn

(5)

The exterior derivative of the k-form

=

Fi1 ,...,ik dxi1 dxik

i1 ,...,ik

is given by
d =

dFi1 ,...,ik dxi1 dxik

(6)

i1 ,...,ik

where dFi1 ,...,ik is computed according to equation (5).

Example 10: Let = (x1 + x23 )dx1 dx2 . Then
d =
=
=
=
=

d(x1 + x23 ) dx1 dx2

dx1 dx1 dx2 + 2x3 dx3 dx1 dx2
2x3 dx3 dx1 dx2
2x3 dx1 dx3 dx2
2x3 dx1 dx2 dx3

Notice the term with two dx1 s is zero, and in the last two steps I just put d in
standard form.
Generalized Stokes Theorem
The generalized version of Stokes Theorem embodies almost everything weve done in
this course; the divergence theorem and the original Stokes Theorem are special cases of
this more general theorem. Before stating this theorem we need a few preliminary remarks.
9

A smooth k-manifold M in lRn may or may not be orientable. If it is orientable then it

has two possible orientations, as we saw in examples above. Its not hard to believe (though
we wont stop to prove) that the boundary M of a k-manifold M is itself a manifold of
dimension k 1. The boundary may or may not be orientable. We will assume that the
manifolds that follow, as well as their boundaries, are orientable. Recall that in the divergence theorem we related a volume integral to a surface integral, and for things to work
out properly we needed the surface (in particular, dA) to be oriented properly. Similarly in
Stokes theorem, after choosing a direction for the unit normal n on a surface S, we then
had to orient S with the right hand rule. Similar considerations hold in higher dimensions.
If we have a manifold M and we have chosen an orientation for M , then this induces a
consistent orientation on M , somewhat akin to the right hand rule. How this can be
done we wont go into for now. The down side is that our integrals may occasionally be o
by a minus sign.
Stokes Theorem: (Generalized version) Let M be a smooth oriented k-manifold M with
consistently oriented smooth boundary M . Let be a k 1 form dened in a neighborhood
of M . Then

d =
.
M

Example 11: Let M be as in Example 5. Given how M was parameterized,

you will probably believe that M can be described as the image of the boundary of the unit disk, u21 + u22 = 1, under the mapping from lR2 to lR4 that
parameterized M . Given that the boundary of the disk can be parameterized by
u1 = cos(t), u2 = sin(t), it seems reasonable that the one dimensional curve M
can be parameterized as
Y(t) = (cos(t), cos(t) sin(t), 3 cos(t) + cos(t) sin(t), 3 sin(t))
(notice Im using Y for M .) Now let = x23 dx1 . Its easy to compute that
d = 2x3 dx1 dx3 .
Lets rst compute

d:

1u2

1
X X
,
)
2 d(
u1 u2
1 1u1
[
]
1 1u2
1
1
0
=
2x3 det
du2 du1

u2 1 u1
1 1u21
1 1u2
1

=
2 2(3 u1 + u1 u2 )u1 du2 du1 = .
2
1 1u1

d =
M

Now for

we have

=
M

(Y (t)) dt

=
0

(3 cos(t) + cos(t) sin(t))2 ( sin(t)) dt

.
2
10

Example 12: Let F = F1 i + F2 j + F3 k be a vector eld in three dimensions

and let = F1 dx1 + F2 dx2 + F3 dx3 , a 1-form. Suppose that M is an orientable
2-manifold (a surface) in three dimensions with one dimensional boundary M .
As we saw in Example 4,

=
F dR.
M

Now lets compute d and integrate it over M . If you write out d you nd that
many terms cancel (they contain a wedge product with identical dxi s) and you
obtain
)

F3 F1
F2 F1

dx1 dx2 +

d =
x1
x2
x1
x3
(
)
F3 F2
+

dx2 dx3
x2
x3

dx1 dx3

Now dene a vector eld G = G1 i + G2 j + G3 k by taking

(

F3 F2
=

x2
x3
(
)
F3 F1
=

x1
x3
)
(
F2 F1

=
.
x1
x2

G1
G2
G3

In short, G = F. Then we can write

d = G1 dx2 dx3 + G2 dx1 dx3 + G3 dx1 dx2 .
If you look at Example 6 (and replace F there by G here), you nd that

d =
M

G dA.

But given that G = F, we conclude from the generalized Stokes Theorem

that

F dR =
( F) dA
M

which is the usual version of Stokes Theorem in three dimensions!

Example 12: Let M be a bounded region in three dimensions with orientable
two dimensional boundary M . Again, let F = F1 i + F2 j + F3 k be a smooth
vector eld dened on M and dene the 2-form
= F1 dx2 dx3 + F2 dx1 dx3 + F3 dx1 dx2
Again referring back to Example 6, we nd that

=
M

11

F dA,

the ux of F over M . Now compute d. Most of the terms are zero and we end
up with
(

d =

F1 F2 F3
+
+
dx1 dx2 dx3 = ( F)dx1 dx2 dx3 .
x1
x2
x3

Now from Example 8 we know that (if M is oriented correctly)

d =
M

F dV.

From the general Stokes Theorem we conclude that

F dA =

which is the divergence theorem!

12

F dV.